Jaspreet Singh
Menu
Highlighted Projects & Accomplishments
I am greatful you have taken the view to view some of my accomplishments and projects. interest in my projects. I have selected some projects that highlight my most recent accomplishments. These accomplishments skim the surface of the depth of what I have learned, especially as it pertains to leadership and development. I urge you to view my CV and my full list of projects to get a more comprehensive idea of the projects and tasks I can do.
​
I also want to reinforce the idea that I truly do love to learn, and I view it as something that should be a human right and is a great priveledge and honor. I hope my work is testament to this belief.
​
​
​
​
01
Motion-Edge Detector
​Using numpy and cv2 in Python, I developed a black and white motion detection camera. The threshold detection can be altered in this as well. The video on the left displays a trial of this project.
​
​Event cameras, such as the Dynamic Vision Sensor (DVS), are bio-inspired vision sensors that output pixel-level brightness changes instead of standard intensity frames. They offer significant advantages over standard cameras, namely a very high dynamic range, no motion blur, and a latency in the order of microseconds
​
​
​
Click here to go to my Github page and view more about this project.
02
MNIST Digit Classification with MLP
In this project, I developed an artificial neural network in Python using numpy, keras, and Tensorflow. I generated a Multi-Layer Perceptron (MLP) to classify images of digits (1-4) from the MNIST dataset, achieving an accuracy of around 95%. I explored the impact of different activation functions (ReLU vs. Sigmoid) and variations in the number of hidden layers and nodes. Through experimentation, I gained insights into balancing model complexity to avoid underfitting and overfitting. I also observed the trade-offs between activation functions, understanding the trade-off for better performance in deeper networks against overfitting without regularization techniques.
​
Click to see the associated project with code on my Github portfolio. ​​​
03
1-D Cochlear Model
In this project, I created a neuromorphic event-based audio sensor that models a 1D cochlea using a microphone. When audio is spoken into the microphone, the sensor processes the audio signals (in this case, spoken digits) by applying Fast Fourier Transform (FFT), breaking down the audio into frequency bands, and generating spike-based events based on power thresholds in each band.
The project mimics the human cochlea, which converts sound into electrical signals by detecting frequency components.
Building a neuromorphic 1D cochlea is important because it simulates how the human ear processes sound in real-time, enabling efficient, low-power signal processing similar to biological systems. This approach is especially valuable for developing auditory prosthetics (e.g., cochlear implants) and low-power audio processing systems for speech recognition and other applications, as it captures key auditory features while minimizing computational and power demands.
​
04
Keyword Spotting Using Spiking Neural Networks
Keyword Spotting, or using A.I to detect spoken words with increasing accuracy has a multitude of uses in today's day and age - beginning with Alexa, to translation, to accessibility. Spilking Neural Networks may offer an alternative approach with lower latency and higher computational power.
In this project, using Loihi's neuromorphic chip platform Lava, I created a Spiking Neural Network to perform keyword spotting for spoken digits (0-9) from a validated dataset. Additionally, Pytorch and keras were used to implement a LIF neuron model to perform keyword spotting.
​
I have not made my code or paper publicly available for privacy reasons. Please reach out to view this code, paper, or presentation and I would be happy to share more information.
05
Medical Device: Rotating Vial Holder
This project focuses on developing a rotating medication vial holder to improve the efficiency and safety of lumbar epidural injections. It addresses a critical clinical need: maintaining sterility while drawing medication from vials without requiring a second medical professional, thereby reducing procedure times and minimizing the risk of needlestick injuries for healthcare staff. The rotating holder is designed to secure vials, provide hands-free operation using a foot pump, and enhance workflow efficiency.
By applying ethnographic studies, the team observed the real-world challenges in clinical settings, particularly the reliance on medical assistants to hold vials, which prolongs procedure times and compromises sterility. The project aims to eliminate the need for an additional assistant, lowering the risks of needle sticks and infections, and reducing human error. The design underwent several iterations, incorporating feedback from clinicians and focusing on ease of use, sterility, and compactness.
The innovation has the potential to significantly impact healthcare workflows, increasing the number of patients seen per day and improving safety in high-risk needle procedures.
06
Regulatory Affairs Consultant for Spinal Cord Stimulator
Links to this project and a more detailed description will be available soon. I appreciate your assistance as this website undergoes construction.
07
Neural Data Analysis
I have completed several projects in Matlab which perform data analysis. Some examples of recent projects have included:
1. Brain-Computer Interface (BCI) Movement Data Analysis (MATLAB)
Project Focus:
-
Neural data analysis using a linear regression model to decode hand movements in monkeys based on brain signals.
Technical Elements:
-
Comet Plot Visualization: Movement trajectories of the monkey’s hand are plotted dynamically to study how it reaches toward targets during a movement task​(
-
Decoder Analysis: Linear regression models were used to predict hand movement based on neural signals. The accuracy of the decoder was around 94%, but there is scope for improvement using cross-validation or more advanced machine learning methods such as neural networks​Time-Lagged Analysis: The impact of various time lags on decoding accuracy was explored by applying delays between brain signal input and hand movement output​(BCI systems v.2).
-
Spike Count Tuning Curves: Neural spike counts were analyzed across different neurons to study target-specific firing activity​
Engineering Relevance:
-
This project involves signal processing, linear modeling, and decoder optimization for real-time applications in brain-computer interfaces. Methods for improving accuracy include adjusting hyperparameters and incorporating neural network-based decoders.
2. Neural Visual Perception Data Analysis (MATLAB)
Project Focus:
-
Simulating visual processing in the retina and V1 cortex using neural data and computational models in MATLAB.
Technical Elements:
-
Mach Bands Illusion: The phenomenon of edge detection is modeled computationally to mimic how retinal ganglion cells detect brightness and contrast changes​(Singh, J Computer Visio…)​(Singh, J Computer Visio…).
-
Convolution for Visual Processing: Receptive fields of retinal ganglion cells were convolved with visual stimuli to simulate neural response to light intensity and edge detection​(Singh, J Computer Visio…)​(Singh, J Computer Visio…).
-
Gabor Function for V1 Simulation: Gabor filters were used to model neurons in the visual cortex (V1), which are sensitive to specific orientations and spatial frequencies. Convolution with images helped detect edges and textures, simulating V1 neuron activity​(Singh, J Computer Visio…).
Engineering Relevance:
-
This analysis integrates image processing, convolution operations, and Gabor filters to simulate biological vision systems. It emphasizes how artificial systems can replicate biological processes for applications like computer vision and neural prosthetics.
3. Reaching Task Data Analysis (MATLAB)
Project Focus:
-
Behavioral and neural analysis of monkeys performing a reaching task with data collected from the premotor cortex.
Technical Elements:
-
Movement and Eye Position Tracking: Analysis of hand and eye movements during a delayed center-out task, with visualizations comparing actual and predicted movement paths​(reaching hw disclaim).
-
Reaction Time and Velocity Analysis: Velocity and reaction time of hand movements were studied to calculate how quickly the monkey responded to stimuli. An ANOVA test was used to assess statistical significance between reaction times across different trials​(reaching hw disclaim).
-
Peristimulus Time Histogram (PSTH): Neural spike data were plotted to show the firing rates of neurons before and after stimulus presentation. Directional selectivity was detected in certain neurons​(reaching hw disclaim).
Engineering Relevance:
-
This project combines biomechanics and neuroscience data analysis to understand motor responses. The statistical methods used, such as PSTH and ANOVA, are critical in quantifying neuron response times and performance in motor tasks.
4. Spike Sorting for Neural Data (MATLAB)
Project Focus:
-
The uploaded .m file likely contains code for spike sorting, a method used to identify and classify action potentials from multi-neuron recordings in neural data.
Technical Elements:
-
Spike sorting involves filtering raw data, detecting spikes, and clustering them based on waveform characteristics. This process is essential for interpreting neural recordings from brain-computer interfaces or neuroscience experiments.
Engineering Relevance:
-
Spike sorting is central to signal processing in neural engineering, where separating signals from different neurons is crucial for accurate data interpretation. This method supports advanced applications in neural prosthetics and neurofeedback systems.
Overall Engineering Themes:
-
Signal Processing: Across all projects, signal processing is a core element, whether decoding movement from neural activity or processing visual stimuli to simulate perception.
-
Statistical Modeling and Machine Learning: Linear regression, decoder models, ANOVA, and potential neural networks are used to model relationships between neural inputs and behavioral outputs.
-
Neuroprosthetics Applications: The projects focus on translating neural activity into actionable outputs like hand movements or visual processing, which are key areas in brain-computer interfaces and prosthetics.
-
Computational Neuroscience Tools: MATLAB is heavily used for modeling neural data and analyzing physiological responses, combining methods from both biomedical engineering and computer vision.
08
Changemaker Scholar
As part of my ongoing journey to become a changemaker scholar through the Big Idea Center, I have committed to exploring innovation and entrepreneurial concepts and what change-making in action and its impact looks like across various disciplines, professions, and areas of interest. This program combines students from a variety of disciplines, and ​​I joined the scholar series inaugural class. ​
09
10
Undergraduate Thesis & Publication
According to the CDC, 61% of adults had at least one adverse event in childhood and 16% had 4 or more types of adverse childhood events. Abuse, neglect, inadequate access to necessary resources, maltreatment, and trauma affect millions of children each day. It is necessary to understand the influences early life experiences have had much later, including in adulthood. My undergraduate thesis explored the neural effects on stress reactivity in adulthood for those who underwent childhood adversity, specifically focusing on perceived socioeconomic standing.
​
The thesis, "Childhood Adversity and Perceived Social Standing: Negative Associations of Trauma and Absence of Effects of Perceived Childhood Socioeconomic Status on Resting-State Functional Connectivity within a Central Visceral Circuit," explores the long-term impacts of adverse childhood experiences (ACEs) on adulthood brain function. Specifically, the research investigates how childhood trauma and socioeconomic deprivation influence resting-state functional connectivity in a neural network central to stress and emotional regulation, including regions like the amygdala, bed nucleus of the stria terminalis (BNST), and subgenual anterior cingulate cortex (sgACC). The findings suggest that trauma leads to lower connectivity between key areas associated with emotional regulation, potentially contributing to the development of affective disorders such as anxiety and depression. However, no significant relationship was found between socioeconomic status or perceived social standing and brain connectivity. These insights contribute to understanding how early life adversity shapes neural circuits linked to mental health outcomes​​
​
​
11
Machine Learning: Sale Prediction Model Using Ridge-Regression
My mother often tells me to plan for the future - which is a handy tool for a business for a busy graduate student. Using ridge regression, I created a model for future predicted sales for Jeep Wrangler. I used a dataset containing car sales of Jeep wranglers, I created a prediction model for future sales of cars. This predictive model is a helpful way to forecast what future trends in a company may look like based on previous data.
​
See the model here.
​
​
12